Kernel Obstruction Theorem: Diagonal Collapse

• Kernel Obstruction Theorem is an abstract diagonal obstruction in the closed implication–falsity fragment, revealing inconsistency when combining generative evaluation with total decision.
• It employs a primitive closure predicate C with structural rules like modus ponens and consistency, alongside evaluation and excluded‐middle completeness, to establish reflective fixed points.
• The theorem underpins reflective paradoxes in logical systems and informs limitations on classifier interfaces in formal, categorical, and computational settings.

Searching arXiv for the primary paper and closely related context. The Kernel Obstruction Theorem, in the sense developed in "Remarks on Primitive Regulation," is an abstract diagonal obstruction for primitive closure predicates on the closed implication–falsity fragment of propositional syntax. It studies a predicate $C:\mathsf{Form}\to\mathsf{Prop}$, read as acceptance by a primitive regulator, together with two structurally distinct completeness principles: evaluation completeness $\mathsf{Eval}(C)$, which is generative, and excluded-middle completeness $\mathsf{LEM}(C)$, which is decisional. The theorem states that these two principles cannot coexist with modus ponens and consistency: $$\mathsf{Eval}(C)\land \mathsf{LEM}(C)\land \mathsf{MP}\land \mathsf{Cons}\Rightarrow\bot$$ (Rosko, 18 May 2026).

1. Formal language and closure kernel

The formal setting is the closed implication–falsity fragment

$A,B ::= \bot \mid A\to B,$

with object-level negation defined by $\neg A \equiv A\to\bot$. The language contains no atoms, variables, quantifiers, or modalities; formulas are finite trees over $\{\bot,\to\}$. A basic syntactic lemma states that no formula is its own negation: $\forall A,\; A\neq \neg A$. This excludes literal fixed points at the level of syntax and forces all fixed-point phenomena to occur internally through acceptance by $C$ (Rosko, 18 May 2026).

A primitive closure predicate is a map $C:\mathsf{Form}\to\mathsf{Prop}$. The intended reading is that $\mathsf{Eval}(C)$0 means “$\mathsf{Eval}(C)$1 is accepted by $\mathsf{Eval}(C)$2.” The predicate is purely propositional and parametric; it is not introduced as a truth predicate or model predicate. From $\mathsf{Eval}(C)$3 one defines the closure equivalence relation

$\mathsf{Eval}(C)$4

This equivalence is the kernel of the classifier in the sense that $\mathsf{Eval}(C)$5 and $\mathsf{Eval}(C)$6 are indistinguishable for $\mathsf{Eval}(C)$7 up to accepted implication in both directions.

Only three structural properties are used. Modus ponens is

$\mathsf{Eval}(C)$8

Consistency is

$\mathsf{Eval}(C)$9

Excluded-middle completeness is

$\mathsf{LEM}(C)$0

A recurrent point in the paper is that no contraction or other object-logical rules are required. A common misunderstanding is therefore ruled out at the outset: the obstruction does not depend on a rich ambient proof theory; it uses only detachment, consistency, and the two completeness interfaces.

2. Generativity, decision, and representability

The first completeness principle is evaluation completeness. One fixes an arbitrary type $\mathsf{LEM}(C)$1 together with an evaluation operation

$\mathsf{LEM}(C)$2

written $\mathsf{LEM}(C)$3. Then

$\mathsf{LEM}(C)$4

This says that every formula-valued behavior of codes is representable by some code, pointwise up to closure equivalence. The paper characterizes this as generativity. A goal-restricted variant also suffices:

$\mathsf{LEM}(C)$5

for fixed goal $\mathsf{LEM}(C)$6.

The second completeness principle is decisional:

$\mathsf{LEM}(C)$7

A stronger Boolean form is

$\mathsf{LEM}(C)$8

It satisfies

$\mathsf{LEM}(C)$9

and therefore implies $$\mathsf{Eval}(C)\land \mathsf{LEM}(C)\land \mathsf{MP}\land \mathsf{Cons}\Rightarrow\bot$$0 by case analysis on $$\mathsf{Eval}(C)\land \mathsf{LEM}(C)\land \mathsf{MP}\land \mathsf{Cons}\Rightarrow\bot$$1.

The paper contrasts this with refutation:

$$\mathsf{Eval}(C)\land \mathsf{LEM}(C)\land \mathsf{MP}\land \mathsf{Cons}\Rightarrow\bot$$2

Unlike $$\mathsf{Eval}(C)\land \mathsf{LEM}(C)\land \mathsf{MP}\land \mathsf{Cons}\Rightarrow\bot$$3, refutation imposes no coverage condition. The always-false classifier inhabits $$\mathsf{Eval}(C)\land \mathsf{LEM}(C)\land \mathsf{MP}\land \mathsf{Cons}\Rightarrow\bot$$4 vacuously. This asymmetry is central: the obstruction targets total classification, not mere partial rejection.

A second common misunderstanding is explicitly blocked by the interface-based formulation. The theorem does not assert that $$\mathsf{Eval}(C)\land \mathsf{LEM}(C)\land \mathsf{MP}\land \mathsf{Cons}\Rightarrow\bot$$5 is universally inhabited for every plausible $$\mathsf{Eval}(C)\land \mathsf{LEM}(C)\land \mathsf{MP}\land \mathsf{Cons}\Rightarrow\bot$$6; it states that if such an evaluation interface is provided, then combining it with total decision has the stated consequence.

3. Reflective fixed points and diagonal collapse

The diagonal mechanism is a Curry-style fixed-point construction internal to the kernel relation. For any goal $$\mathsf{Eval}(C)\land \mathsf{LEM}(C)\land \mathsf{MP}\land \mathsf{Cons}\Rightarrow\bot$$7, a formula $$\mathsf{Eval}(C)\land \mathsf{LEM}(C)\land \mathsf{MP}\land \mathsf{Cons}\Rightarrow\bot$$8 is called a Curry fixed point at $$\mathsf{Eval}(C)\land \mathsf{LEM}(C)\land \mathsf{MP}\land \mathsf{Cons}\Rightarrow\bot$$9 if

$A,B ::= \bot \mid A\to B,$0

Under $A,B ::= \bot \mid A\to B,$1, for every goal $A,B ::= \bot \mid A\to B,$2 there exists such a $A,B ::= \bot \mid A\to B,$3. The proof is by self-application: instantiate $A,B ::= \bot \mid A\to B,$4 at

$A,B ::= \bot \mid A\to B,$5

obtain $A,B ::= \bot \mid A\to B,$6 with $A,B ::= \bot \mid A\to B,$7 for all $A,B ::= \bot \mid A\to B,$8, set $A,B ::= \bot \mid A\to B,$9, and specialize at $\neg A \equiv A\to\bot$0. This yields

$\neg A \equiv A\to\bot$1

For $\neg A \equiv A\to\bot$2, one obtains a negation fixed point

$\neg A \equiv A\to\bot$3

Thus $\neg A \equiv A\to\bot$4 implies $\neg A \equiv A\to\bot$5 (Rosko, 18 May 2026).

Once such a $\neg A \equiv A\to\bot$6 exists, excluded-middle completeness forces classification of it. The branch-collapse lemma states that if $\neg A \equiv A\to\bot$7 holds and $\neg A \equiv A\to\bot$8, then either branch yields $\neg A \equiv A\to\bot$9. If $\{\bot,\to\}$0, then $\{\bot,\to\}$1 and modus ponens give $\{\bot,\to\}$2; from $\{\bot,\to\}$3 and $\{\bot,\to\}$4 one obtains $\{\bot,\to\}$5. If $\{\bot,\to\}$6, then $\{\bot,\to\}$7 and modus ponens give $\{\bot,\to\}$8, and the same detachment yields $\{\bot,\to\}$9.

This produces the closure-level diagonal collapse:

$\forall A,\; A\neq \neg A$0

Consistency then converts internal collapse into contradiction:

$\forall A,\; A\neq \neg A$1

The paper presents this as the Kernel Obstruction Theorem in formal summary form:

$\forall A,\; A\neq \neg A$2

$\forall A,\; A\neq \neg A$3

$\forall A,\; A\neq \neg A$4

$\forall A,\; A\neq \neg A$5

and therefore

$\forall A,\; A\neq \neg A$6

4. Why the obstruction is a kernel obstruction

The term “kernel” is tied to the equivalence relation $\forall A,\; A\neq \neg A$7. The relation says that $\forall A,\; A\neq \neg A$8 accepts arrows both ways between two formulas, so classification by $\forall A,\; A\neq \neg A$9 factors through the quotient $C$0. In the paper’s categorical gloss, $C$1 is the kernel congruence induced by the classifier’s accepted implications. The obstruction is therefore not merely a diagonal paradox in syntax; it is a failure of a classifier to consistently decide formulas that collapse in its own kernel class.

The fixed point produced by $C$2 is reflective rather than syntactic. Since no formula is its own negation, the identity $C$3 never arises as a tree equality. What occurs is the weaker but decisive statement

$C$4

That is enough because $C$5 demands that $C$6 decide one side or the other, while the accepted kernel arrows guarantee that either decision propagates to falsity through modus ponens.

The paper explicitly places this mechanism alongside broader diagonal patterns. It states that $C$7 functions as point-surjectivity up to $C$8, in the spirit of Lawvere’s fixed-point theorem; that the result matches Rice-style and undefinability obstructions for nontrivial properties over reflective domains; and that the function $C$9 carries the self-reference. These comparisons situate the theorem within a familiar family of reflective impossibility arguments, while the actual proof remains purely propositional and interface-driven (Rosko, 18 May 2026).

A plausible implication is that the theorem isolates a minimal diagonal core. The ambient syntax is austere, the logical rules are sparse, and equality is relaxed to closure equivalence. Yet this already suffices for inconsistency once generativity and total decision are combined.

5. Strengthenings, variants, and scope

The theorem strengthens immediately to Boolean decision. Since $C:\mathsf{Form}\to\mathsf{Prop}$0, one obtains

$C:\mathsf{Form}\to\mathsf{Prop}$1

Even without $C:\mathsf{Form}\to\mathsf{Prop}$2, the combination of $C:\mathsf{Form}\to\mathsf{Prop}$3, $C:\mathsf{Form}\to\mathsf{Prop}$4, and $C:\mathsf{Form}\to\mathsf{Prop}$5 yields internal collapse to $C:\mathsf{Form}\to\mathsf{Prop}$6. The added contradiction requires consistency to discharge $C:\mathsf{Form}\to\mathsf{Prop}$7 into $C:\mathsf{Form}\to\mathsf{Prop}$8.

By contrast, refutation is not obstructed. The paper defines

$C:\mathsf{Form}\to\mathsf{Prop}$9

and notes that the always-false classifier inhabits this specification vacuously. Because no global coverage requirement is imposed, no diagonal forcing occurs.

The theorem is parametric in $\mathsf{Eval}(C)$00, and the paper lists several informal classes of examples. One may take $\mathsf{Eval}(C)$01 as a provability or derivability closure; one may read $\mathsf{Eval}(C)$02 as an acceptance or truth-like classifier on the fragment; or one may interpret $\mathsf{Eval}(C)$03 as a normalization or definability property. In each case, the applicability depends on whether an evaluation frame witnessing $\mathsf{Eval}(C)$04 is actually provided. The theorem therefore constrains interfaces, not all conceivable predicates outright.

The scope conditions are intentionally tight. The result relies only on the closed implication–falsity fragment, $\mathsf{Eval}(C)$05, $\mathsf{Eval}(C)$06, $\mathsf{Eval}(C)$07, and the bundled evaluation interface. No contraction or additional rules are used. The mechanization also shows that the frame with equality is impossible, and that closure equivalence is the minimal relaxation permitting a reflective fixed point. This directly corrects another possible misconception: the theorem is not a statement about ordinary syntactic fixed points, but about fixed points that live inside the classifier’s kernel.

6. Mechanization and terminological range

The proof is mechanized in Rocq. Formulas are implemented as closed $\mathsf{Eval}(C)$08 trees in module M001, and the evaluation frame provides ceval_apply : Code → Code → Form together with cevaluation_complete, the formal version of $\mathsf{Eval}(C)$09$\mathsf{Eval}(C)$10$\mathsf{Eval}(C)$11\mathsf{Eval}(C)\Rightarrow \exists B.\; B\simeq_C\neg B$\mathsf{Eval}(C)$12C(\bot)$\mathsf{Eval}(C)$13\mathsf{Eval}(C)\land\mathsf{MP}\land\mathsf{LEM}$\mathsf{Eval}(C)$14\bot$\mathsf{Eval}(C)$15\mathsf{Eval}(C)\land\mathsf{MP}\land\mathsf{Cons}\land\mathsf{LEM}$\mathsf{Eval}(C)$16\mathsf{Dec}$; and refutation_not_obstructed, showing that refutation is inhabited by always_false_regulator_refutation. The bundled certificate aporetic_lemma_qed inhabits APORETIC_LEMMA_CONTRACT with empty assumption report (Rosko, 18 May 2026).

The phrase “Kernel Obstruction Theorem” is not unique to this logical setting. In other arXiv contexts it names unrelated results, including a theorem on local Euler obstruction for determinantal varieties (Zhang, 2017), an identification of Boardman’s whole-plane obstruction group with the kernel of a colim–lim interchange morphism in Cartan–Eilenberg systems (Helle et al., 2018), and obstruction-set characterizations in parameterized complexity and kernelization (Fellows et al., 2013). This suggests that the label is field-local rather than canonical. In the present logical usage, however, it specifically denotes the incompatibility of a generative evaluation interface with total excluded-middle decision for a primitive closure predicate over the implication–falsity fragment.